Loading...
Loading
Overview
Registration Activity
Neuron registration trends
This Interval
0
— burn · — PoW
Target / Interval
Loading
Neuron registration trends
Variant calling—detecting genetic mutations in sequencing data—is computationally intensive and tool performance varies dramatically across configurations. Research and clinical pipelines need reliable benchmarks but lack access to distributed compute and automated scoring at scale. A single operator cannot afford to test thousands of caller configurations, and no existing system rewards accurate configuration discovery.
Minos solves this by creating a decentralized marketplace where miners submit variant-calling configurations (GATK, DeepVariant, or BCFtools) and validators independently re-run each config against benchmark data, scoring the output VCF files for accuracy. Validators use hap.py to compare calls against ground truth, computing F1-score, recall, precision, and false-positive counts. Weights are allocated based on round-only scoring—miners earn TAO proportional to detection accuracy, incentivizing them to tune configurations that maximize SNP and indel precision without over-calling.
On Bittensor subnet 107, this workflow is fully decentralized: miners never submit VCF output (only configs), validators cannot trust miner claims (they rescore everything), platform aggregates results from multiple validators for quorum consensus, and canonical tiebreaks ensure fair weight allocation when scores tie. Eligibility gates (recent valid scored rounds), per-coldkey deduplication, and pruning-dust weighting prevent gaming and concentrate rewards on sustained high performance.
Use cases were auto-generated from the GitHub repository and may not reflect the latest changes.
Minos subnet 107 is a Bittensor genomics variant calling subnet where miners submit configurations for genomic variant callers (GATK, DeepVariant, BCFtools) to detect genetic variants in sequencing data. Validators re-run submitted configurations, score the resulting VCF files against ground truth using hap.py benchmarking, and allocate weights based on F1-score accuracy across SNP and indel detection, with eligibility gates and reward normalization mechanisms.
Minos uses a platform-mediated round-based architecture. Miners poll the platform for active rounds (open submission window), download BAM + region, run their variant-calling config (GATK/DeepVariant/BCFtools), submit the config JSON. Validators poll for rounds in scoring phase, fetch submissions, re-run each miner's config locally, score outputs with hap.py against truth VCF, and submit scores to platform.
Platform aggregates scores from multiple validators (quorum-based), provides canonical ranking for tiebreaks, and supplies backfill scores to cover gap miners. Validators then set chain weights (UIDs indexed) according to round-only winner-heavy pruning dust policy with per-coldkey normalization, conditional on eligibility. Emissions follow Bittensor's standard mechanism (TAO rebase) with a burn hotkey as sink.
Config commitment (on-chain digest of config) and paid resubmissions (escalating TAO fee ladder per coldkey) are optional platform features.
This information was auto-extracted from the GitHub repository and may be incomplete or out of date. Always refer to the subnet's GitHub repository for the latest instructions.
Scoring: Validators re-run miner configs on round BAMs, score outputs with hap.py (SNP/indel precision/recall/F1), compute AdvancedScorer (v1 or v2 formula), and record combined_final (0-1 range). Eligibility: Miners must have 5+ valid scored rounds in recent 20 to receive weight/emissions. Weighting: Winner-heavy (top 1-3 miners get 60-80% of budget), pruning dust (remove zero-scorers), per-coldkey normalization (one reward slot per coldkey owner).
Backfill: Secondary validators score gap miners, platform aggregates and distributes peer scores. Tiebreaking: Canonical ranking (quorum validator consensus) for close calls; local fallback (score → submission time → hotkey). Reward formula: burn_rate%, winner_weight%, dust_top_n with dust_decay exponential decay.
Payment: One free config submission per hotkey per round; additional resubmissions cost escalating TAO.
The sections above were auto-generated from the GitHub repository and may not reflect the latest changes. View repository
The Foundational Layer of Genomics
FAQs were auto-generated from the GitHub repository and may not reflect the latest changes.
Holder & total-staked metrics as of 2h ago (hourly snapshot)
GPU not required but optional for DeepVariant. DeepVariant requires 16GB RAM minimum. GATK and BCFtools run on 8GB. SSD storage required with minimum 3000 IOPS (recommended 10000). Docker mandatory (v24.0+). Apple Silicon Macs require Rosetta 2 for x86_64 containers.
GPU not required. Validators must run hap.py for scoring (CPU-intensive). SSD storage strongly recommended with minimum 5000 IOPS (recommended 15000). Docker mandatory (v24.0+). 24/7 uptime required for consistent round scoring. Validators re-run all miner tool configurations so all Docker images required (~15GB total).
Clone repository and install dependencies (Python 3.10+, Bittensor 10.3.x, pysam, numpy, pydantic)
bash install.sh
Verify miner prerequisites: Python version, Docker, Git, system packages
bash scripts/verify.sh --miner
Pre-pull required Docker images (samtools, and one of: gatk, deepvariant, bcftools)
docker pull quay.io/biocontainers/samtools:1.20--h50ea8bc_0 && docker pull broadinstitute/gatk:4.5.0.0
Run demo mode first: one-shot pipeline validation against a fixed chr20 sample, no wallet/chain needed
export MINER_TEMPLATE=gatk && bash start-miner.sh --demo
Run practice mode: score your config against fully-answered chr18-chr22 samples with exact validator scoring, no chain/TAO
export MINER_TEMPLATE=gatk && bash start-miner.sh --practice
Create/attach Bittensor wallet and register hotkey on subnet 107
btcli subnets register --netuid 107 --wallet.name <name> --wallet.hotkey <hotkey>
Start live mining: polls platform for rounds, downloads BAM, runs variant calling, submits config
export MINER_TEMPLATE=gatk && bash start-miner.sh
Optionally supervise miner with PM2 for persistence and monitoring(optional)
bash pm2-miner.sh
Optional: set up OpenClaw + Ditto for local AI assistant help with troubleshooting(optional)
bash scripts/setup_ai_assistant.sh --with-ditto --openclaw
Clone repository and install dependencies (Python 3.10+, Bittensor 10.3.x, pysam, numpy)
bash install.sh
Verify validator prerequisites: Python, Docker, sufficient CPU/RAM/disk
bash scripts/verify.sh --validator
Pre-pull all required Docker images: hap.py, GATK, DeepVariant, BCFtools, samtools (validators re-run all miner configs)
docker pull genonet/hap-py@sha256:03acabe84bbfba35f5a7234129d524c563f5657e1f21150a2ea2797f8e6d05f2 && docker pull broadinstitute/gatk:4.5.0.0 && docker pull google/deepvariant:1.5.0 && docker pull quay.io/biocontainers/bcftools:1.20--h8b25389_0
Create/attach Bittensor wallet and register validator hotkey on subnet 107
btcli subnets register --netuid 107 --wallet.name <name> --wallet.hotkey <hotkey>
Start validator: polls platform for scoring rounds, downloads round data, scores miner submissions, sets weights
export PLATFORM_URL=https://api.theminos.ai && bash start-validator.sh
Optionally supervise validator with PM2 for persistence and monitoring(optional)
bash pm2-validator.sh
